Efficient evaluation of photodynamic therapy on tumor based on deep learning

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cancer Research

Background:

• Photodynamic therapy (PDT) is a non-invasive tumor treatment utilizing photosensitizers and laser irradiation to generate reactive oxygen species, leading to tumor cell death.
• Current methods for assessing PDT efficacy rely on manual live/dead cell counting, which is labor-intensive, time-consuming, and susceptible to variations in dye quality.

Purpose of the Study:

• To develop and validate an automated method for quantifying live and dead cells following PDT treatment.
• To establish a more efficient and objective approach for evaluating the effectiveness of photodynamic therapy.

Main Methods:

• Construction of a dedicated dataset of cells subjected to PDT treatment.
• Training and implementation of the YOLOv3 (You Only Look Once version 3) real-time AI object detection algorithm for cell counting.
• Performance evaluation using mean average precision (mAP) for both live and dead cell detection.

Main Results:

• The YOLOv3 model achieved high accuracy in cell detection, with a mean average precision (mAP) of 94% for live cells and 71.3% for dead cells.
• The AI-driven approach demonstrated efficient and accurate quantification of cell viability post-PDT.

Conclusions:

• The developed AI-based cell counting method provides a significant improvement over traditional manual counting techniques for PDT efficacy assessment.
• This automated approach accelerates the evaluation of PDT treatments, thereby facilitating faster development of novel cancer therapies.